To verify that this metric provides robust model comparisons, we also performed auxiliary analyses in which we used 10-fold cross-validation: specifically, for each left-out fold in turn, we estimated mixing weights using the other 9 folds and evaluated adjusted R2 for PRS computed using these weights on the left-out fold. We then computed average adjusted R2 across the 10 folds. (When analyzing data from an unbalanced case-control study with #cases ≪ #controls, we used stratified 10-fold cross-validation, selecting the folds such that each fold had the same case-control ratio; this applies only to the South Asian UK Biobank T2D analysis.)
